Karhunen-Loeve Based Estimation of Subcarrier Interference Matrices for OFDM over High-Doppler-Spread Doubly-Dispersive Channels

We propose a novel method for estimating intercarrier interference (ICI) matrices for orthogonal frequency division multiplexing (OFDM) on doubly-dispersive (DD) channels using the Karhunen-Loeve transform (KLT). While previous work has described KLT-based and basis-expansion-model (BEM) estimation of DD channel impulse responses h, this work derives the relationship between the autocorrelation matrices R h and RH of h and its corresponding ICI matrix H. When h has N h independent lags, we show that the diagonals of H are stationary even if the columns of h are non-stationary, so that RH is Hermitian. For the considered time-varying Jakes widesense-stationary-uniform-scattering (WSSUS) channel, it is known that rank(R h) increases linearly with the normalized Doppler frequency f d T , requiring a large number of basis functions as f d T increases. By contrast, we show that rank(RH) = N h , the channel length, giving substantial complexity reductions over time-domain (TD) BEM channel estimation at the high Doppler spreads (f d T up to 0.02) considered here. We design a system for joint KLTbased estimation of H and turbo-equalization of OFDM over DD channels. The system employs a MAP-based channel equalizer with a convolutional error correction code; channel estimation and bit error rate (BER) are improved by iterating between the channel estimator, equalizer, and channel decoder. We evaluate the system's BER for different f d T values. We compare against orthogonal matching pursuit (OMP), a common method used for estimating sparse channels, and with a previous KLT-based approach that estimates h. Simulations show that the proposed approach significantly outperforms OMP in terms of BER, and TD-based KLT estimation in terms of mean squared error (MSE).